Composites Science and Technology 171 (2019) 44–53

Contents lists available at ScienceDirect

Composites Science and Technology

journal homepage: www.elsevier.com/locate/compscitech

Critical parameters of carbon nanotube reinforced composites for structural T

health monitoring applications: Empirical results versus theoretical
predictions
Xoan F. Sánchez-Romatea,b,∗, Joaquín Artigasb, Alberto Jiménez-Suárezb, María Sánchezb,
Alfredo Güemesa, Alejandro Ureñab
a
Department of Aerospace Materials and Processes, Escuela Técnica Superior de Ingenieros Aeronáuticos, Universidad Politécnica de Madrid, Plaza del Cardenal Cisneros
3, 28040, Madrid, Spain
b
Materials Science and Engineering Area, Escuela Superior de Ciencias Experimentales y Tecnología, Universidad Rey Juan Carlos, Calle Tulipán s/n, 28933, Móstoles,
Madrid, Spain

A R T I C LE I N FO A B S T R A C T

Keywords: This paper reports on an investigation of the critical parameters which determine the electrical and electro-
(A) Carbon nanotubes mechanical properties of carbon nanotube (CNT) nanocomposites. For this purpose, a novel analytical model,
(B) Electrical properties based on the tunnelling mechanisms of CNTs, is proposed. Three dispersion parameters are introduced in the
(C) Modelling model to reflect the CNT aggregation state. Microscopy analysis and electrical and strain monitoring tests were
Dispersion
carried out on CNT nanocomposites manufactured by toroidal stirring and three roll milling. It is observed that
SHM
electrical conductivity is greatly affected by dispersion procedure as well as strain sensitivity, measured by the
gauge factor (GF). Generally, well dispersed materials have higher conductivities and GF. In this regard, the
aggregate ratio has a prevalent effect. Experimental data and theoretical predictions allow the correlation of
dispersion parameters given by manufacturing procedures with electrical properties to develop highly sensitive
nanocomposites. This demonstrates the potential and applicability of the proposed model.

1. Introduction electrically conductive, is a crucial point in investigations of CNT na-

The use of carbon nanotubes (CNTs) has increased significantly
since the studies of Iijima et al. [1] as they exhibit excellent mechanical
and electrical properties [2–6]. Indeed, tensile strengths of
∼ 11 − 100 GPa , Young's Modulus of ∼ 0.27 − 1 TPa [5,6], and elec-
trical conductivities of 103 − 106S/m have been achieved [7,8]. Of
particular interest is the effect of their addition as fillers in a polymer
matrix on electrical properties. Their high aspect ratio, with values of
more than 1000, makes it possible that very small amounts of CNTs
(below 1 wt %) cause an increase in electrical conductivity of several
orders of magnitude [9–13]. This is because CNTs create a conductive
network inside the matrix, providing electrical pathways. That, in
combination with CNT piezoresistive behaviour, makes possible their
use in strain-sensing applications, which are now attracting the interest
of many researchers [14–17].
In this context, the determination of the percolation threshold, i.e.
the critical volume fraction of nanofiller at which the material becomes
nocomposite electrical behaviour [18–20]. Percolation threshold has
been found to be correlated with CNT geometry and orientation [21], as
well as dispersion procedure, for CNT content ranging from values of
0.01 wt % [18] to above 0.5 wt % [9,22]. For this reason, the influence
of these parameters has been widely studied and theoretical models for
predicting the percolation threshold and electrical conductivity of CNT
nanocomposites have been proposed [8,23–26].
The strain monitoring capabilities of CNT-based composites are
greatly influenced by percolation threshold and electrical conductivity
and correlations between these parameters have also been determined
by theoretical modelling [27–29]. Tunnelling resistance has been found
to be the dominant mechanism that rules the electromechanical beha-
viour of CNT nanocomposites [30] and is given by the distance between
two adjacent CNTs. Experimental and numerical investigations have
correlated tunnelling distance with CNT content [31,32], showing that
the highest sensitivities are generally achieved at CNT contents near the
percolation threshold. However, the influence of dispersion techniques

∗
Corresponding author. Department of Aerospace Materials and Processes, Escuela Técnica Superior de Ingenieros Aeronáuticos, Universidad Politécnica de
Madrid, Plaza del Cardenal Cisneros 3, 28040, Madrid, Spain.
E-mail address: xoan.fernandez.sanchezromate@urjc.es (X.F. Sánchez-Romate).

https://doi.org/10.1016/j.compscitech.2018.12.010
Received 5 September 2018; Received in revised form 2 December 2018; Accepted 12 December 2018
Available online 13 December 2018
0266-3538/ © 2018 Elsevier Ltd. All rights reserved.
X.F. Sánchez-Romate et al.
– and thus the CNT dispersion state – on the electromechanical cap-
abilities of CNT nanocomposites remains to be investigated.
In a previous study, an improved analytical model based on ag-
gregation parameters was proposed [33], analysing different dispersion
techniques and CNT fillers. This approach gives a precise idea of how
exactly the dispersion state and CNT geometry affect electrical con-
ductivity, but it does not allow prediction of the electromechanical
properties of CNT-based nanocomposites as it does not take into ac-
count tunnelling mechanisms. For this reason, the aim of this study is to
examine the influence of dispersion state on the electromechanical
behaviour of CNT nanocomposites.
To achieve this, electrical conductivity measurements, strain mon-
itoring tests, and dispersion state characterization by optical micro-
scopy were carried out using different dispersion conditions and CNT
content. A novel approach based on CNT dispersion state is proposed
and critical parameters that influence the electromechanical properties
are identified and correlated. The validity and applicability of this
model are demonstrated by comparing theoretical predictions to ex-
perimental data.
2. Materials and methods
2.1. Materials
Nanocomposites were manufactured with multiwall carbon nano-
tubes (MWCNTs) embedded in an epoxy resin. Commercially available
Nanocyl NC3150 MWCNTs, with an average diameter of 9.5 nm and a
length up to 1 μm, were sourced. The resin was a low-viscosity DGBEA
Huntsman Araldite LY556 cured with aromatic amine Araldite XB3473.
2.2. Manufacturing of CNT nanocomposites
MWCNT nanocomposites were manufactured in three steps: addi-
tion of the nanoparticles into the matrix and mechanical dispersion,
degasification of the mixture, and curing.
Two dispersion procedures were used to prepare the MWCNT-epoxy
mixtures: three roll milling and a combination of toroidal stirring and
three roll milling. Toroidal stirring was carried out using a Dispermat
dissolver at 6000 rpm for 15 min. The three roll milling dispersion
process was optimized in previous studies [34,35]. It consists of a
progressive reduction in the gap between the rolls, keeping the velocity
of the third (last) roll constant at 250 rpm. The parameters of the three
roll milling dispersion procedure are shown in Table 1. To provide a
more detailed analysis of the effect of dispersion procedure on electrical
properties, different CNT content mixtures (0.05, 0.1 and 0.3 wt %)
were manufactured in different numbers of three roll milling cycles: 1,
3 and 7 for 0.05 and 0.1 wt %, with and without toroidal stirring; 1, 3
and 7 for 0.3 wt % mixtures without toroidal stirring, as it is not pos-
sible to achieve a 3D toroidal flow when adding such a high content).
Table 2 summarizes the different conditions tested.
After the dispersion process, a degasification step was carried out to
evacuate the air entrapped in the epoxy mixture. For this purpose, the
mixture was heated to 80 °C and then degasification was done under
vacuum for 15 min. After that, hardener was added to the mixture in a
proportion 100:23 monomer:hardener. Finally, doped resins were cured
in an oven at 140 °C for 8 h in a 270 × 120 × 4 mm3 mould.
Table 1
Process conditions for three roll milling dispersion.
Cycle First roll gap (μm) Last roll gap (μm)
1 120 40
2 75 25
3 45 15
4–7 15 5
45
Composites Science and Technology 171 (2019) 44–53
Table 2
Processing methods and conditions for CNT-epoxy mixture manufacturing.
Content (% wt.) Designation Procedure
0.05/0.1/0.3 C1 Three roll milling process. Cycle 1.
0.05/0.1/0.3 C3 Three roll milling process. Cycles 1–3.
0.05/0.1/0.3 C7 Three roll milling process. Cycles 1–7.
0.05/0.1 Disp Toroidal stirring at 6000 rpm for 15 min
0.05/0.1 DC1 Toroidal stirring + C1
0.05/0.1 DC3 Toroidal stirring + C3
0.05/0.1 DC7 Toroidal stirring + C7
2.3. Characterization of nanocomposites
CNT dispersion in the CNT-epoxy mixtures was analysed using a
Leica high transmission optical microscope (TOM), equipped with a
Nikon Coolpix 990 camera. In addition, to analyse CNT distribution and
agglomerate morphology in the nanocomposite, materials were frac-
tured under cryogenic conditions. First, a pre-crack treatment was done
with the nanocomposite immersed in liquid N2 for 5 min. Then, the
sample was fractured and coated with platinum (5.6 nm). The micro-
structural study was carried out by Field Emission Gun Scanning
Electron Microscopy FEG-SEM using a Nova NanoSEM FEI 230 appa-
ratus from Philips.
DC volume conductivity was evaluated according to ASTM D257
using a Source Measurement Unit (SMU, Keithley Instrument Inc. mod.
2410). Three specimens (10 × 10 × 1 mm ) were tested for each condi-
tion. Silver paint was used to ensure good contact between the elec-
trode and the specimen, the surface of which was treated by brushing.
The voltage applied was within the range of 0–20 V and electrical re-
sistance was determined by the slope of the I-V curve.
2.4. Strain monitoring tests
Tensile tests were carried out according to ASTM D638 at a test rate
of 5 mm/min and electrical resistance was measured using a multimeter
(Agilent 34410A). The electrodes were made of copper wire with silver
paint to ensure good contact with the nanocomposite surface. Adhesive
tape was used to protect the electrodes from environmental issues. The
electrodes were placed 30 mm apart on the central region of the tensile
specimen, subjected to a uniform strain. Tensile specimens were iso-
lated from the machine by using and adhesive layer at the grips. Fig. 1
shows an image of the electrode placement.
3. Results and discussion
Electrical conductivity and strain monitoring tests were carried out
for the specimens manufactured by three roll milling and a combination
of three roll milling and toroidal stirring. An analytical model has been
proposed to identify the main parameters affecting the electrical
properties of CNT nanocomposites and a comparison between theore-
tical predictions and experimental measurements has been established.
Fig. 1. Electrode placement on tensile specimen.
X.F. Sánchez-Romate et al.
Fig. 2. Electrical conductivity measurements for the different CNT nano-
composites prepared by three roll milling (solid lines) and the combination of
toroidal stirring and three roll milling (dashed lines).
3.1. Experimental results
Figs. 2 and 3 summarize the electrical properties of CNT/epoxy
nanocomposites under tested conditions. At the lower counts of three
roll mill processing, i.e. for C1 to C3 for all CNTs, electrical conductivity
increases drastically as shown in Fig. 2 (solid lines). This is due to the
effective breakage of larger agglomerates, caused by induced shear
forces, which leads to a better homogenization of the CNT, as observed
by comparing TOM images after one to three cycles of the three roll
milling process (Fig. 4(a) and (b)). In addition, as stated in a previous
study, these shear forces also induce a stretching effect on CNTs [33],
which causes an increase in the effective aspect ratio and thus a higher
conductivity as it induces a decrease of the percolation threshold [23]. .
On the other hand, the increase in electrical conductivity is less sig-
nificant because of the greater CNT breakage effect [36,37] for C3 to
C7. Moreover, the three roll milling effect on electrical conductivity at
high cycles is less significant with an increasing CNT content, even
leading to a slight reduction on electrical conductivity from the 3-step
to 7-step three roll milling process. This effect can be explained by
Fig. 3. Measured GF for the different CNT nanocomposites prepared by three
roll milling (solid lines) and a combination of toroidal stirring and three roll
milling (dashed lines).
46
Composites Science and Technology 171 (2019) 44–53
increased interactions between carbon nanotubes, which lead to a
higher CNT breakage effect due to a decrease in their effective elastic
modulus [38,39].
On the other hand, an enhancement in electrical conductivity is
observed with a previous toroidal stirring step when comparing to three
roll milling process, specially at lower three roll milling steps (i.e. C1 to
DC1 and C3 to DC3 of Fig. 2). This could be explained by the 3D flow
induced by toroidal stirring leading to a higher homogenization effect
[40] (Fig. 4(c)) without a CNT breakage as the shear forces involved are
much lower. This causes the mean aggregate size to be reduced and
thus CNT interactions are also lower when passing through the three
roll mill, resulting in a well dispersed mixture (Fig. 4(d)). However, no
significant differences between lower and higher cycles of three roll
milling are observed with a prior toroidal stirring process i.e. com-
paring C3 with DC3, and C7 and DC7. This could be because the main
homogenization effect takes place during the toroidal stirring step.
Thus, in this case, three roll milling does not have a relevant effect on
CNT dispersion, resulting in very similar dispersion states when com-
paring DC1 to DC7 (Fig. 4(d) and (e)). In addition, the effect of toroidal
stirring is less evident at lower CNT content.
Fig. 3 summarizes the gauge factor (GF), defined as the variation of
normalized resistance between applied strain achieved, that is ΔR / R0 ,
ε
for the different specimens manufactured at low strain levels
(ε = 0.005) . A significant increase in sensitivity is observed with three
roll mill steps due to the high disaggregation effect induced by shear
forces. On the other hand, toroidal stirring improves sensitivity when
combined with a few cycles of the three roll mill process, but at higher
cycles it does not seem to have a significant effect and could even give
rise to lower GF, The main reason can lie in the fact that at higher three
roll milling steps, there is a more prevalent CNT breakage effect due to
the shear forces involved leading, thus, to a lower electrical con-
ductivity and GF. In addition, GF values at different CNT contents were
similar, although a slightly higher sensitivity is achieved with lower
CNT contents, especially in the three roll milling process. This could be
due to a similar dispersion state observed at different CNT contents
(Fig. 4(e) and (f)).
Therefore, from the experimental measurements, it is clear that
there is a correlation between the dispersion procedure and electro-
mechanical properties. Thus, to attain a deeper understanding of elec-
tromechanical properties, an analytical model correlating the crucial
parameters is proposed.
3.2. Theoretical analysis
In the model, CNT agglomerates are considered to be perfect
spheres, as shown in Fig. 5, supposing that adjacent carbon nanotubes
in an agglomerate are in contact. Electrical resistance is thus composed
of three terms, CNT intrinsic, contact and tunnelling resistance, as
follows:
RT = RCNT + R contact + Rtunnel (1)
The electrical intrinsic resistance of CNTs can be obtained using the
following expression:
1 l
RCNT = ·
σCNT A (2)
where l is the length between two contact points, A the cross-sectional
area and σCNT the intrinsic conductivity of the CNT.
For contact resistance, an expression such as that stated by Gong
et al. [41] can be used, assuming there is no deformation at CNT
junctions:
h 1
R contact = ·
e 2 MT (3)
where M is the number of conducting channels, estimated between 400
and 500 for MWCNTs as stated previously by Li et al. [42]; h2 is the
e
X.F. Sánchez-Romate et al. Composites Science and Technology 171 (2019) 44–53

Fig. 4. TOM images of CNT dispersion for (a) C1, (b) C3, (c) Disp, (d) DC1, (e) DC3 at 0.05 wt % and (f) DC7 at 0.1 wt % (red circles indicate the presence of larger
aggregates). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Planck resistance and T the transmission probability for electrons tun- following expression:
nelling through the polymer, which is estimated as:
1 l h 1
R cluster = RCNT + R contact = · + 2·
σCNT A e MT (5)
⎛ −d w ⎞
T = exp ⎜ ⎟
h / 8mφ (4) The expression presented in Eq. (5) denotes the electrical resistance
⎝ CNT ⎠
term, which is assumed to be invariable with strain and is called in-
where d w is the van der Waals distance, 0.34 nm, m the mass of an trinsic resistance. It is considered invariable as it refers to the intrinsic
electron and φCNT the work function of CNTs (set at 4.5 eV) [43]. resistance of the cluster. CNTs can be deformed due to strain and it
Thus for a CNT cluster, total resistance can be calculated by the sum induces a change on the electrical conductivity. However, due to the 3D
of intrinsic and contact resistance at CNT junctions, leading to the disposition of CNTs in a cluster, Poison effects can mitigate the

Fig. 5. Schematic of the proposed model showing (left) the real dispersion state and (right) the equivalent block disposition.

47
X.F. Sánchez-Romate et al. Composites Science and Technology 171 (2019) 44–53

electrical change due to CNT deformation. In addition, the poor CNT- resistance due to applied strain as a function of the dispersion state:
matrix interaction in an agglomerate also mitigates the overall nano- Ra Rd
composite strain effect over the CNTs and therefore, for simplicity, ΔR (ξd R a + ξa Rd ) R a R d ⎛ ξd R a0 + ξa R d 0 ⎞
= R a0 R d 0
−1= ·⎜ ⎟ − 1
intrinsic resistance will be considered invariable. R0 R a0 R d 0 ⎝ ξd R a + ξa R d ⎠
(ξd R a0 + ξa Rd0) (10)
However, the other term, the tunnelling resistance, changes with
applied strain and gives the electromechanical properties of the nano- in which the subscript 0 denotes the initial stage with no applied strain.
composite. This expression can be simplified by defining several non-dimensional
The tunnelling resistance is calculated using the following expres- parameters, such as: the ratio of agglomerated ( χa ) and dispersed ( χd )
sion, as stated by Hu et al. [27]: CNTs; the coefficients of electrical resistance before and after applied
R R
strain ( χa = R a and χd = R d respectively); the ratio between ag-
h2t 4πt a0 d0
Rtunnel = exp ⎛ 2mΦ ⎞ glomerated and dispersed CNTs (φ = ξa/ ξd , also defined as the ag-
Ae 2 2mφ ⎝ h ⎠ (6)
gregate ratio); the correlation between Rd0 and Ra0 (Ψ). This leads to the
in which A is the tunnelling area, estimated as the cross-sectional area following expression:
d 2
of the CNTs (that is A = π· CNT
4
); t the tunnelling distance, given by the
ΔR 1 + φΨ ⎞
distance between two neighbouring CNTs in which the tunnelling effect = χa χd ·⎛⎜ ⎟ − 1
R0 ⎝ a + φχd Ψ ⎠
χ (11)
occurs; and Φ the height of the barrier of epoxy (set at 0.75 eV) [27,29].
In the case of well dispersed material, that is φ = 0 , Eq. (11) leads to
3.2.1. Modelling of the dispersion state ΔR
R0
= χd − 1, and if the CNTs are all in the form of aggregates, that is
To date, most studies have assumed that CNTs are perfectly dis- ΔR
φ → ∞, it leads to R = χa − 1.
persed in the matrix and thus electromechanical behaviour is directly 0
In addition, the electrical conductivity of CNT nanocomposites can
given by the expression in Eq. (6). However, the dispersion state is a key
be calculated from R using the following expression:
factor for electrical conductivity and electromechanical behaviour, as
shown in the experimental measurements, so it is necessary to take it L Leq 1 1 (ξd Ra + ξa Rd )
σ= = = = ·
into account when modelling the strain monitoring capabilities of CNT R·A R·Leq2 R·Leq Leq Ra Rd (12)
nanocomposites.
For this reason, and to attain a better understanding of CNT dis- where Leq is the length of the equivalent cubic element, defined in a
persion state, three parameters are defined: ξa is the volume fraction of previous study [33].
material that is occupied by CNT agglomerates, i.e. the agglomerate Therefore, it is possible to correlate electrical properties with the
volume fraction dominated by intrinsic resistance; ξd is the volume CNT dispersion state in a very simple way. However, it is important to
fraction occupied by well dispersed CNTs in which tunnelling resistance point out that this proposed model is only valid at lower strain levels, in
is prevalent and over the percolation threshold; ξnon is the volume which the effect of the breakage of electrical pathways due to matrix
fraction of the material in which the CNT concentration is much lower fracture and damage [44] is not so prevalent. For higher applied strain,
than the average content, called the non-percolated region. it is necessary to take these effects into account in the model to obtain
more accurate results.
Va Vd Vn
ξa = ξd = ξn =
V V V (7)
3.2.2. Modelling approach to inter-particle distance
where Va , Vd and Vn are the volumes of the agglomerated, dispersed and To calculate the tunnelling resistance, it is necessary to determine
non-percolated regions respectively. the mean tunnelling distance, also called the inter-particle distance
In a previous study [23], ε has been defined as the localized volume (IPD). This is defined as the mean distance of two adjacent particles. IPD
of CNT in an agglomerate using the following expression: depends fundamentally on CNT volume content, as it implies a higher
nVCNT number of nanoparticles in the material. The main issue is to determine
ε= (1 > ε > ε0) this value for any given content.
πD3
6 (8) IPD depends on the number of particles per unit volume, n, fol-
lowing a power law of exponent α [31]:
where ε0 is defined as the minimum localized volume of CNTs needed to
consider them as agglomerates, n is the number of nanoparticles in a 1 α
IPD ∼ ⎛ ⎞
cluster and D the diameter of the equivalent sphere. n
⎝ ⎠ (13)
Once the volume fraction of agglomerated and well dispersed CNTs
Therefore, supposing that the value of the tunnelling distance at the
is defined, it is necessary to correlate them to the electrical properties of
percolation threshold, t0, is known and set at 1.4 nm, as proposed by Li
the materials. As previously stated, the electrical resistance of ag-
et al. [30], it is possible to obtain the mean tunnelling distance, t, for
gregates, Ra , is given by the intrinsic resistance, Rintrinsic , using Eq. (5).
any CNT concentration using the following expression:
In the case of well dispersed CNTs, the electrical resistance of the dis-
α
persed regions, Rd , is given by Rtunnel , defined in Eq. (6), while the re- ρ
sistance of the non-percolated region can be considered infinite. t = t0 ⎜⎛ 0 ⎟⎞
⎝ρ⎠ (14)
Therefore, the material can be divided into three different blocks (ξa , ξd
and ξnon ), i.e. the volume fraction of agglomerated, well dispersed and where ρ and ρ0 are the volume fraction at a general concentration and
non-percolated nanoparticles as shown in Fig. 5. This configuration the percolation threshold, respectively. The percolation threshold is
overestimates the total electrical resistance of the nanocomposite but determined using the following expression, assuming CNTs are straight
gives an idea of how dispersion state affects the electrical properties of and with no alignment:
the material. Thus, it is possible to calculate the equivalent electrical
3 3 1
resistance along the material, R. ρ0 = ·
4 Λ2 (15)
1 1 1 1 Ra Rd
= ξa· + ξd · + ξnon· →R= in which Λ is the aspect ratio of CNTs. As stated previously, three roll
R Ra Rd  R ∞ (ξ d a + ξa R d )
R
milling has a breakage effect on CNTs, so it affects the value of ρ0 due to
∼0 (9)
reduction of the effective aspect ratio of CNTs. The resulting aspect
Using this expression, it is possible to see the change in electrical ratio after a three roll milling process can be calculated based on the

48
X.F. Sánchez-Romate et al. Composites Science and Technology 171 (2019) 44–53

formula proposed by Yamamoto et al. [45]: In addition, IPD is observed to have a combined effect on the sen-
sitivity of the material. At lower values of IPD, the sensitivity increases
ηγ˙ ln(2Λ) − 1.75
∼ 100 with tunnelling distance because of a relevant effect of tunnelling re-
E 2Λ4 (16)
sistance which varies exponentially with applied strain. However, at a
where E is the Young's Modulus of CNTs, (1 TPa); η the viscosity of the certain IPD, sensitivity begins to decrease again. This is because the
epoxy resin, (4 Pa·s) and γ̇ the shear rate of the three roll milling pro- value of tunnelling resistance is so high that electricity trends to go
cess. through agglomerates, even when the CNTs are well dispersed.
The exponent α can be calculated assuming that, at the maximum In this regard, Fig. 6(b) shows the correlation between IPD and CNT
CNT content ( ρm ) , IPD is equal to the minimum distance between two geometries at different contents based on Eq. (11). It can be seen that
particles, i.e. the van der Waals distance (d w ), resulting in the following lower aspect ratios lead to higher IPDs as the percolation threshold is
expression: increased. In addition, IPD decreases with CNT content due to the
d presence of a higher number of particles per unit volume.
α ln ⎛ wd ⎞
t0 ( )
ρ0
ρm
= dw → α = ⎝ t0 ⎠
ρ
ln ⎛ 0 ⎞
⎜ ⎟
The correlation between electrical conductivity and aggregation
parameters is shown in Fig. 7 at different CNT concentrations. At lower
⎝ ρm ⎠
π CNT content, the electrical conductivity is given mainly by the value of
ρm = 6
(max . package factor of a sphere ) (17) ξa due to the presence of preferential pathways along the CNT ag-
glomerates. For a value of ξa , it is observed that the presence of more
As previously stated, the effect of tunnelling resistance is relevant
well dispersed material leads to higher conductivities. At higher CNT
when CNTs are well dispersed, so to calculate the mean tunnelling
content, variation of electrical conductivity with ξd is more significant,
distance, only the volume fraction of well dispersed CNTs is taken into
as the IPD of well dispersed CNTs is lower, and thus tunnelling re-
account. Assuming the localized volume content of well dispersed CNTs
sistance is also lower, so its contribution to electrical conductivity is
is equal to the overall CNT volume fraction in the material, Eq. (14) can
more prevalent.
be rewritten as follows:
Therefore, the above parametric study gives an accurate idea of how
α
ρ dispersion state can affect electromechanical properties. It has been
td = t0 ⎛⎜ 0 ⎞⎟
⎝ρ⎠ (18) shown that the highest conductivities are achieved for well dispersed
materials, i.e. those with a lower aggregate ratio. In this context,
electrical conductivity is significantly dependent on ξa at low CNT
3.3. Parametric study content, and the effect of the well dispersed region, ξd , is controlling at
high CNT content, while the IPD distance is greatly affected by CNT
A parametric study was carried out to evaluate the effect that dis- aspect ratio and content. Thus, it is necessary to compare theoretical
persion state has on the electromechanical sensitivity of the material. predictions to the experimental data to provide a detailed analysis of
To achieve this purpose, the change in contact resistance due to the the influence of dispersion technique on the electromechanical prop-
variation of tunnelling distance on the junctions was considered neg- erties of CNT/epoxy nanocomposites.
ligible. Thus, Ra was constant with applied strain, which implies that
χa = 1. In addition, χd can be calculated at a known tunnelling distance
and thus so can the ratio between the electrical resistance of agglom- 3.4. Comparison between theoretical approach and experimental
erates and well dispersed CNTs, Ψ. measurements
Fig. 6(a) shows the calculated GF from Eq. (11) with varying IPD
and aggregation rate, (φ) at a value of ε = 0.005. It can be seen that the Fig. 8 presents the effect of the dispersion procedure at different
better the dispersion state, that is the lower the aggregate ratio, the mean IPD on the GF predicted by Eq. (11), with varying aggregation
higher the sensitivity of the material. This is because there are fewer parameters, compared to experimental data obtained from the strain
preferential electrical pathways the lower the agglomerates size, so the monitoring tests. The values of GF used for the calculation are those at
whole material acts as a sensor. Therefore, the highest values of ag- lower strain levels, i.e. ε = 0.005. In addition, the effective aspect ratio
gregation rate lead to the lowest sensitivities as those agglomerates act was calculated using Eq. (16). To account for the higher interactions
as preferential areas since the intrinsic resistance is usually much lower between CNTs at higher contents, the resulting aspect ratio was mul-
than tunnelling resistance. tiplied by a factor of 1.2, 1 and 0.7 for 0.05, 0.1 and 0.3 wt %

Fig. 6. (a) Effect of the aggregate ratio and IPD on GF and (b) effect of the aspect ratio and CNT content on the mean IPD.

49
X.F. Sánchez-Romate et al.
Fig. 7. Effect of aggregation parameters on electrical conductivity at (a) 0.1 (b) 0.2 and (c) 0.5 vol % CNT.
Fig. 8. Effect of different process conditions on the aggregation ratio and IPD.
Red, black and blue symbols refer 0.05, 0.1 and 0.3 wt %, respectively. (For
interpretation of the references to colour in this figure legend, the reader is
referred to the Web version of this article.)
50
Composites Science and Technology 171 (2019) 44–53
respectively. These factors were set to take into account the interaction
forces between adjacent nanoparticles. As previously stated [37], the
expression used in Eq. (16) overestimates the real aspect ratio at high
CNT contents, while at low contents the shear forces presented are
much lower.
The aggregate ratio was observed to change dramatically depending
on CNT content and dispersion procedure. A general increase is ob-
served with CNT content from values from 0.024 to 0.13 at 0.05 wt % to
0.14–4 at 0.3 wt %. This is due to a higher number of nanoparticles
forming larger agglomerates and increasing the volume fraction of CNT
aggregates. In addition, the three roll milling process was observed to
reduce drastically the aggregate ratio by approximately an order of
magnitude. However, the addition of a toroidal stirring step does not
seem significantly to improve the aggregate ratio of CNT nanocompo-
sites over a three roll milling process, as previously stated, because
toroidal stirring does not significantly improve homogeneity as an ad-
ditional step to a three roll milling process. However, a slight en-
hancement is observed at 0.1 wt % due to a better homogenization ef-
fect of toroidal stirring at higher CNT content. IPD is also greatly
affected by process conditions. IPD generally decreases with CNT con-
tent from 1.02 to 1.24 nm at 0.3 wt % to 1.44 nm at 0.05 wt %.
Nevertheless, it increases with three roll milling cycles, due to reduc-
tion in the effective aspect ratio, which induces an increase in the
percolation threshold. This effect is more pronounced at higher con-
tents because breakage – and thus the reduction of the effective aspect
ratio – of CNTs is much higher.
Fig. 9 summarizes the main dispersion parameters obtained for each
X.F. Sánchez-Romate et al.
Fig. 9. Calculated dispersion parameters for each condition. Solid lines indicate
the non-percolated area and dashed lines the aggregate ratio. The green-co-
loured areas correspond to well dispersed materials and the red regions to
poorly dispersed ones. Red, black and blue symbols denote 0.05, 0.1 and 0.3 wt
%, respectively. (For interpretation of the references to colour in this figure
legend, the reader is referred to the Web version of this article.)
process condition. ξa generally increases with CNT content (from
0.0005 to 0.004 at 0.05 wt % to 0.04–0.055 at 0.3 wt %) and the three
roll milling and toroidal stirring process (varying by an order of mag-
nitude in some cases). This seems contradictory to expected behaviour.
However, higher values of ξa do not mean higher entanglement. It is
correlated with the fraction of total volume occupied by agglomerates.
The degree of entanglement is given by the aggregate ratio. Therefore,
higher values of ξa in combination with lower values of φ indicate
higher agglomerated areas with lower localized volume content, i.e.
better homogenization of CNT dispersion, as shown in Fig. 10. In this
regard, it is observed that the non-percolated area decreases sig-
nificantly with three roll milling cycles, from values of 0.98–0.99 to
0.7–0.87 from one to seven cycles of three roll milling, respectively; and
with the addition of a toroidal stirring step, reaching values below 0.1,
indicating the homogenization effect that these processes have on CNT
dispersion by inducing the creation of more electrical pathways, which
leads to an increase in well dispersed regions. Toroidal stirring also
induces a decrease in the non-percolated region to values around
0.75–0.85 due to the homogenization effect on CNT dispersion. In ad-
dition, a general decrease is also observed with CNT content, especially
at C7, from 0.87 at 0.05 wt % to 0.7 at 0.3 wt %, due to a higher
breakage effect from three roll milling.
In addition, by observing the fractured surface in the FEG-SEM
images of Fig. 11, the agglomerated areas, occupying a small volume
fraction of the material, show a higher concentrated distribution of
CNTs at low steps of the three roll milling process (Fig. 11 (a)) in
comparison to high-step three roll milling, in which well dispersed
Fig. 10. Schematics of dispersion state for (left) low ξa and high φ, (center) medium ξa and φ and (right) high ξa and low φ.
51
Composites Science and Technology 171 (2019) 44–53
regions are more common (Fig. 11 (b) and (c)). This means that CNTs
are more entangled in the case of low steps and there is a significant
decrease in the non-percolated area with the number of three roll
milling steps, as stated in the theoretical analysis. A similar CNT dis-
tribution for high-step three roll milling is observed compared with
toroidal stirring (Fig. 11 (d)), as expected.
Therefore, a good agreement between theoretical predictions and
experimental measurements is found. It is important to point out that
the determination of dispersion parameters, such as the agglomerate
volume fraction or non-percolated area, is not trivial using microscopy
analysis. In addition, the dispersion state could change during curing
due to a re-aggregation of agglomerates [46,47]. However, the results
found in the literature for a more detailed microscopy analysis [48–50]
also show good agreement with predictions based on this novel ana-
lytical model, indicating an increase in well dispersed areas with the
addition of proper dispersion procedures.
4. Conclusions
Experimental measurements and theoretical studies were carried
out to determine the crucial parameters that define electrical con-
ductivity and – especially – the strain monitoring properties of CNT/
epoxy nanocomposites. To achieve this, CNT nanocomposites were
produced by two dispersion techniques under varying process condi-
tions and their electrical conductivity and strain monitoring capabilities
were determined while the CNT dispersion state was observed using
transmitted optical microscopy. In addition, a novel analytical model
has been proposed to calculate electrical conductivity and electro-
mechanical properties. This model uses three descriptive parameters of
the aggregation state: the aggregate and dispersed volume fractions and
the non-percolated areas.
From the experimental measurements, it was found that electrical
conductivity increases drastically from one to three cycles of three roll
milling. The addition of a toroidal stirring step does not cause sig-
nificant changes at seven cycles of three roll milling. The highest sen-
sitivities were achieved with seven cycle three roll milling at lower
contents due to a significant enhancement of CNT dispersion. However,
at higher contents, the effect of three roll milling process can be ne-
gative due to a more prevalent CNT breakage effect induced by higher
CNT interactions. It is also observed that toroidal stirring does not
drastically improve the sensitivity obtained with three roll milling.
From the proposed model, it was found that electrical conductivity
increases with agglomerate volume fraction and, given the same ag-
gregate fraction, is improved with higher fractions of well dispersed
CNTs, which explains the positive effect of three roll milling on elec-
trical conductivity. Furthermore, the aggregate ratio was found to be
the main parameter determining the sensitivity capabilities of CNT
nanocomposites. Highest sensitivities were achieved for well dispersed
materials in which there were no preferential electrical pathways.
Moreover, inter-particle distance was also found to be a relevant
parameter for electrical properties and a simple, but effective, corre-
lation with CNT content was proposed. It was pointed out that there is
an optimal value of inter-particle distance at which sensitivity is the
highest, depending on the aggregate ratio. Thus, from these results, it is
X.F. Sánchez-Romate et al.
Fig. 11. FEG-SEM images of fractured surface at (a) C1, (b) C3, (c) C7 and (d) DC7 0.05 wt % CNT. Red areas indicate the presence of aggregates. (For interpretation
of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
possible to select the optimum process conditions to achieve the highest
sensitivities.
To conclude, experimental measurements and theoretical predic-
tions were compared and showed good agreement. Thus, the potential
and applicability of the analytical model proposed was demonstrated.
Acknowledgements
This work was supported by the Ministerio de Economía y
Competitividad of Spanish Government [Project MAT2013-46695-C3-
1-R] and Comunidad de Madrid Government [Project P2013/MIT-
2862]. The authors wish to thank Proof-Reading-Service.com for
editing a version of this manuscript.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.compscitech.2018.12.010.
References
[1] S. Iijima, Helical microtubules of graphitic carbon, Nature 354 (6348) (1991)
56–58.
[2] R.S. Ruoff, D.C. Lorents, Mechanical and thermal properties of carbon nanotubes,
Carbon 33 (7) (1995) 925–930.
[3] E.W. Wong, P.E. Sheehan, C.M. Lieber, Nanobeam mechanics: elasticity, strength,
and toughness of nanorods and nanotubes, Science 277 (5334) (1997) 1971–1975.
[4] S. Xie, W. Li, Z. Pan, B. Chang, L. Sun, Mechanical and physical properties on carbon
nanotube, J. Phys. Chem. Solid. 61 (7) (2000) 1153–1158.
[5] M.F. Yu, O. Lourie, M.J. Dyer, K. Moloni, T.F. Kelly, R.S. Ruoff, Strength and
breaking mechanism of multiwalled carbon nanotubes under tensile load, Science
287 (5453) (2000) 637–640.
52
Composites Science and Technology 171 (2019) 44–53
[6] B. Peng, M. Locascio, P. Zapol, S. Li, S.L. Mielke, G.C. Schatz, H.D. Espinosa,
Measurements of near-ultimate strength for multiwalled carbon nanotubes and ir-
radiation-induced crosslinking improvements, Nat. Nanotechnol. 3 (10) (2008)
626–631.
[7] W. Deheer, W. Bacsa, A. Chatelain, T. Gerfin, R. Humphreybaker, L. Forro,
D. Ugarte, Aligned carbon nanotube films - production and optical and electronic-
properties, Science 268 (5212) (1995) 845–847.
[8] W. Bao, S. Meguid, Z. Zhu, M. Meguid, Modeling electrical conductivities of na-
nocomposites with aligned carbon nanotubes, Nanotechnology 22 (48) (2011)
485704.
[9] A. Allaoui, S. Bai, H. Cheng, J. Bai, Mechanical and electrical properties of a
MWNT/epoxy composite, Compos. Sci. Technol. 62 (15) (2002) 1993–1998.
[10] C. Martin, J. Sandler, M. Shaffer, M. Schwarz, W. Bauhofer, K. Schulte, A. Windle,
Formation of percolating networks in multi-wall carbon-nanotube–epoxy compo-
sites, Compos. Sci. Technol. 64 (15) (2004) 2309–2316.
[11] Y.J. Kim, T.S. Shin, H. Do Choi, J.H. Kwon, Y. Chung, H.G. Yoon, Electrical con-
ductivity of chemically modified multiwalled carbon nanotube/epoxy composites,
Carbon 43 (1) (2005) 23–30.
[12] M.B. Bryning, M.F. Islam, J.M. Kikkawa, A.G. Yodh, Very low conductivity
threshold in bulk isotropic single‐walled carbon nanotube–epoxy composites, Adv.
Mater. 17 (9) (2005) 1186–1191.
[13] N. Hu, Z. Masuda, C. Yan, G. Yamamoto, H. Fukunaga, T. Hashida, The electrical
properties of polymer nanocomposites with carbon nanotube fillers,
Nanotechnology 19 (21) (2008) 215701.
[14] E.T. Thostenson, T. Chou, Carbon nanotube networks: sensing of distributed strain
and damage for life prediction and self healing, Adv. Mater. 18 (21) (2006)
2837–+.
[15] C. Li, E.T. Thostenson, T. Chou, Sensors and actuators based on carbon nanotubes
and their composites: a review, Compos. Sci. Technol. 68 (6) (2008) 1227–1249.
[16] N. Hu, Y. Karube, M. Arai, T. Watanabe, C. Yan, Y. Li, Y. Liu, H. Fukunaga,
Investigation on sensitivity of a polymer/carbon nanotube composite strain sensor,
Carbon 48 (3) (2010) 680–687.
[17] N.J. Heeder, A. Shukla, V. Chalivendra, S. Yang, K. Park, Electrical response of
carbon nanotube reinforced nanocomposites under static and dynamic loading, Exp.
Mech. 52 (3) (2012) 315–322.
[18] J. Sandler, J. Kirk, I. Kinloch, M. Shaffer, A. Windle, Ultra-low electrical percolation
threshold in carbon-nanotube-epoxy composites, Polymer 44 (19) (2003)
5893–5899.
[19] J.Z. Kovacs, B.S. Velagala, K. Schulte, W. Bauhofer, Two percolation thresholds in
X.F. Sánchez-Romate et al.
carbon nanotube epoxy composites, Compos. Sci. Technol. 67 (5) (2007) 922–928.
[20] W. Bauhofer, J.Z. Kovacs, A review and analysis of electrical percolation in carbon
nanotube polymer composites, Compos. Sci. Technol. 69 (10) (2009) 1486–1498.
[21] S. Kale, F.A. Sabet, I. Jasiuk, M. Ostoja-Starzewski, Effect of filler alignment on
percolation in polymer nanocomposites using tunneling-percolation model, J. Appl.
Phys. 120 (4) (2016) 045105.
[22] S. Cui, R. Canet, A. Derre, M. Couzi, P. Delhaes, Characterization of multiwall
carbon nanotubes and influence of surfactant in the nanocomposite processing,
Carbon 41 (4) (2003) 797–809.
[23] J. Li, P.C. Ma, W.S. Chow, C.K. To, B.Z. Tang, J. Kim, Correlations between per-
colation threshold, dispersion state, and aspect ratio of carbon nanotubes, Adv.
Funct. Mater. 17 (16) (2007) 3207–3215.
[24] W. Bao, S. Meguid, Z. Zhu, Y. Pan, G. Weng, A novel approach to predict the
electrical conductivity of multifunctional nanocomposites, Mech. Mater. 46 (2012)
129–138.
[25] Y. Wang, G.J. Weng, S.A. Meguid, A.M. Hamouda, A continuum model with a
percolation threshold and tunneling-assisted interfacial conductivity for carbon
nanotube-based nanocomposites, J. Appl. Phys. 115 (19) (2014) 193706.
[26] J. Li, M. Östling, Conductivity scaling in supercritical percolation of nanoparti-
cles–not a power law, Nanoscale 7 (8) (2015) 3424–3428.
[27] N. Hu, Y. Karube, C. Yan, Z. Masuda, H. Fukunaga, Tunneling effect in a polymer/
carbon nanotube nanocomposite strain sensor, Acta Mater. 56 (13) (2008)
2929–2936.
[28] Y. Kuronuma, T. Takeda, Y. Shindo, F. Narita, Z. Wei, Electrical resistance-based
strain sensing in carbon nanotube/polymer composites under tension: analytical
modeling and experiments, Compos. Sci. Technol. 72 (14) (2012) 1678–1682.
[29] B. De Vivo, P. Lamberti, G. Spinelli, V. Tucci, L. Vertuccio, V. Vittoria, Simulation
and experimental characterization of polymer/carbon nanotubes composites for
strain sensor applications, J. Appl. Phys. 116 (5) (2014) 054307.
[30] C. Li, E.T. Thostenson, T. Chou, Dominant role of tunneling resistance in the elec-
trical conductivity of carbon nanotube-based composites, Appl. Phys. Lett. 91 (22)
(2007) 223114.
[31] T. Takeda, Y. Shindo, Y. Kuronuma, F. Narita, Modeling and characterization of the
electrical conductivity of carbon nanotube-based polymer composites, Polymer 52
(17) (2011) 3852–3856.
[32] R. Rahman, P. Servati, Effects of inter-tube distance and alignment on tunnelling
resistance and strain sensitivity of nanotube/polymer composite films,
Nanotechnology 23 (5) (2012) 055703.
[33] X.F. Sanchez-Romate, A. Jimenez-Suarez, M. Sanchez, A. Guemes, A. Urena, Novel
approach to percolation threshold on electrical conductivity of carbon nanotube
reinforced nanocomposites, RSC Adv. 6 (49) (2016) 43418–43428.
[34] M. Sanchez, J. Rams, M. Campo, A. Jimenez-Suarez, A. Urena, Characterization of
carbon nanofiber/epoxy nanocomposites by the nanoindentation technique,
Compos. Part B Eng. 42 (4) (2011) 638–644.
[35] A. Jimenez-Suarez, M. Campo, M. Sanchez, C. Romon, A. Urena, Influence of the
functionalization of carbon nanotubes on calendering dispersion effectiveness in a
53
Composites Science and Technology 171 (2019) 44–53
low viscosity resin for VARIM processes, Compos. Part B Eng. 43 (8) (2012)
3482–3490.
[36] S. Fu, Z. Chen, S. Hong, C.C. Han, The reduction of carbon nanotube (CNT) length
during the manufacture of CNT/polymer composites and a method to simulta-
neously determine the resulting CNT and interfacial strengths, Carbon 47 (14)
(2009) 3192–3200.
[37] J. Kovacs, W. Bauhofer, The influence of the dispersion quality of carbon nanotubes
in an epoxy matrix on the electrical conductivity of the composite, Proceeding of
the 17th International Conference on Composite Materials, 2009.
[38] D. Shi, X. Feng, Y.Y. Huang, K. Hwang, H. Gao, The effect of nanotube waviness and
agglomeration on the elastic property of carbon nanotube-reinforced composites, J.
Eng. Mater. Technol. 126 (3) (2004) 250–257.
[39] C.S. Jarali, S.R. Basavaraddi, B. Kiefer, S.C. Pilli, Y.C. Lu, Modeling of the effective
elastic properties of multifunctional carbon nanocomposites due to agglomeration
of straight circular carbon nanotubes in a polymer matrix, J. Appl. Mech. 81 (2)
(2014) 021010.
[40] A.K. Chakraborty, T. Plyhm, M. Barbezat, A. Necola, G.P. Terrasi, Carbon nanotube
(CNT)–epoxy nanocomposites: a systematic investigation of CNT dispersion, J.
Nanoparticle Res. 13 (12) (2011) 6493–6506.
[41] S. Gong, Z. Zhu, E. Haddad, Modeling electrical conductivity of nanocomposites by
considering carbon nanotube deformation at nanotube junctions, J. Appl. Phys. 114
(7) (2013) 074303.
[42] H. Li, W. Lu, J. Li, X. Bai, C. Gu, Multichannel ballistic transport in multiwall carbon
nanotubes, Phys. Rev. Lett. 95 (8) (2005) 086601.
[43] R. Gao, Z. Pan, Z. Wang, Work function at the tips of multiwalled carbon nanotubes,
Appl. Phys. Lett. 78 (12) (2001) 1757–1759.
[44] Y. Shindo, Y. Kuronuma, T. Takeda, F. Narita, S. Fu, Electrical resistance change
and crack behavior in carbon nanotube/polymer composites under tensile loading,
Compos. Part B Eng. 43 (1) (2012) 39–43.
[45] S. Yamamoto, T. Matsuoka, Dynamic simulation of flow-induced fiber fracture,
Polym. Eng. Sci. 35 (12) (1995) 1022–1030.
[46] M. Chapartegui, N. Markaide, S. Florez, C. Elizetxea, M. Fernandez, A. Santamaria,
Curing of epoxy/carbon nanotubes physical networks, Polym. Eng. Sci. 52 (3)
(2012) 663–670.
[47] M. Aravand, S.V. Lomov, I. Verpoest, L. Gorbatikh, Evolution of carbon nanotube
dispersion in preparation of epoxy-based composites: from a masterbatch to a na-
nocomposite, Express Polym. Lett. 8 (8) (2014) 596–608.
[48] S.R. Bakshi, R.G. Batista, A. Agarwal, Quantification of carbon nanotube distribu-
tion and property correlation in nanocomposites, Compos. Part A Appl. Sci. Manuf.
40 (8) (2009) 1311–1318.
[49] I.H. Sul, J.R. Youn, Y.S. Song, Quantitative dispersion evaluation of carbon nano-
tubes using a new analysis protocol, Carbon 49 (4) (2011) 1473–1478.
[50] B.M. Tyson, R.K.A. Al-Rub, A. Yazdanbakhsh, Z. Grasley, A quantitative method for
analyzing the dispersion and agglomeration of nano-particles in composite mate-
rials, Compos. B Eng. 42 (6) (2011) 1395–1403.